From Protein Structure to Function with Bioinformatics.pdf

128 P. Tompa5.5 Prediction of the Function of IDPsAs suggested by the foregoing considerations, reliable all-round prediction ofthe functions of IDPs is still a long way off, and we have only taken the firststeps towards this goal. As discussed in the next section, there are severalapproaches that may shed some light on the function of an IDP not yet experimentallycharacterized. Functional correlation of the global pattern on disorder(Lobley et al. 2007), sequence-based prediction of short LMs by a variety ofalgorithms (Davey et al. 2006; Neduva and Russell 2006), prediction of MoRFsin IDPs/IDRs (Mohan et al. 2006; Vacic et al. 2007), and combination ofsequence information with disorder (Iakoucheva et al. 2004; Radivojac et al.2006) are reasonable approaches to assess the function of an unknown piece ofdisordered protein.5.5.1 Correlation of Disorder Pattern and FunctionJones and colleagues have taken a direct approach to find association between theglobal pattern of disorder and the function of a protein (Lobley et al. 2007) describedby standard Gene Ontology (GO) categories. It was first found that both location- andlength-descriptors of disorder correlate with functional categories associated withsignal transduction and transcription regulation. Both molecular function (MF) andbiological process (BP) annotations were used. The location descriptors displayedseveral trends associated with GO categories, such as an elevated level in the middleof the protein in transcription regulator, DNA binding, and RNA pol II transcriptionfactor functions, in the C-terminus in transcription factor activator, transcription factorrepressor, and transcription factor or in the N-terminus in potassium channelannotated proteins. Length descriptors showed even more significant associationswith function than position descriptors. For example, disordered regions of more than500 continuous residues are over-represented in transcription-related categories,whereas shorter regions of the order of 50 residues or fewer are overrepresented inproteins performing metal ion binding, ion channel, and GTPase regulatory functions.The observed associations could be used to improve prediction of protein function:an SVM predictor applied to 26 GO categories, prediction of 11 BP categories and12 MF categories showed improvements resulting from the addition of disorder features.In all, disorder adds significantly to the prediction of protein function, withmore significant improvements observed in BP than in MF classification.5.5.2 Predicting Short Recognition Motifs in IDRsA completely different but relevant approach is to predict short sequence motifsin IDPs/IDRs, which may then be directly related to certain functions, such aspost-translational modification or binding to cognate partners. As suggested